A Speech Activity Adaptation System to Improve Human-Machine Communication

A machine which adapts to the speech of any human speaker and responds at an optimal utterance speed would be useful for the smooth exchange of information in human-machine communication, and the present study is concerned with the development of such a system. The author has already proposed a method for adapting machine conversational speed to speaker utterance speed. Based on sensory evaluation, this paper demonstrates that the speech activity adaptation itself is more essential for the smooth exchange of information than the adaptation to speech speed. Also, by analyzing the characteristics of speech activity estimation error as a function of observation time (talkspurt-silence period), the minimal number of talkspurt-silence periods required to estimate speech activity is determined. Finally, a conversational messaging system has been developed, which plays back stored speech segments controlled by the speech activity adaptation algorithm.